I'm wondering did the kernel folk ever go back and review the split of the Megaraid driver into two? I understand why these kinds of changes sometimes need to happen, but seems to me that it caused a lot of practical issues.

I'm still left with machines where I (allegedly) have to manually patch the kernel sources to get the right Megaraid driver loaded (presumably because no one had a complete list of hardware IDs), and today I was upgrading an ancient system (to make a test server) and I needed to manually recreate the initrd image with the legacy Megaraid driver in (because it was omitted by default, and the root file system was thus not found). This kind of thing scares off administrators, and Google is little help unless you understand the topic as it mostly lists thousands of others having similar issues.

Meanwhile my bitter experiences of Adaptec, and LSI PC Hardware RAID solutions leads me to be jaundiced about hardware RAID on Linux. Sometimes a good UPS software RAID and keeping stupid people away from your hardware is a better solution. Waiting for disks to spin around is a stupid way of committing data to permanent storage, and the sooner something better becomes truly mainstream the better. Flash storage is pretty much there, but I think administrators need to be clear how much they need this. Our servers are almost universally limited by committing writes (except for the big hardware RAID one, where I can't diagnose what is making it slow because the kernel says there is no I/O at all, but I suspect it is Postgres writing too much too often (because I know how the software is written)).

Wise word follow...
http://linux.yyz.us/why-software-raid.html